4 * Copyright (c) 2003-2008 Fabrice Bellard
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
25 #include "qemu-common.h"
26 #include "qemu/timer.h"
27 #include "qemu/sockets.h" // struct in_addr needed for libslirp.h
28 #include "sysemu/qtest.h"
29 #include "slirp/libslirp.h"
30 #include "qemu/main-loop.h"
31 #include "block/aio.h"
33 #include "sysemu/hax.h"
37 #include "qemu/compatfd.h"
39 /* If we have signalfd, we mask out the signals we want to handle and then
40 * use signalfd to listen for them. We rely on whatever the current signal
41 * handler is to dispatch the signals when we receive them.
43 static void sigfd_handler(void *opaque)
45 int fd = (intptr_t)opaque;
46 struct qemu_signalfd_siginfo info;
47 struct sigaction action;
52 len = read(fd, &info, sizeof(info));
53 } while (len == -1 && errno == EINTR);
55 if (len == -1 && errno == EAGAIN) {
59 if (len != sizeof(info)) {
60 printf("read from sigfd returned %zd: %m\n", len);
64 sigaction(info.ssi_signo, NULL, &action);
65 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
66 action.sa_sigaction(info.ssi_signo,
67 (siginfo_t *)&info, NULL);
68 } else if (action.sa_handler) {
69 action.sa_handler(info.ssi_signo);
74 static int qemu_signal_init(void)
80 * SIG_IPI must be blocked in the main thread and must not be caught
81 * by sigwait() in the signal thread. Otherwise, the cpu thread will
82 * not catch it reliably.
85 sigaddset(&set, SIG_IPI);
86 sigaddset(&set, SIGIO);
87 sigaddset(&set, SIGALRM);
88 sigaddset(&set, SIGBUS);
89 /* SIGINT cannot be handled via signalfd, so that ^C can be used
90 * to interrupt QEMU when it is being run under gdb. SIGHUP and
91 * SIGTERM are also handled asynchronously, even though it is not
92 * strictly necessary, because they use the same handler as SIGINT.
94 pthread_sigmask(SIG_BLOCK, &set, NULL);
96 sigdelset(&set, SIG_IPI);
97 sigfd = qemu_signalfd(&set);
99 fprintf(stderr, "failed to create signalfd\n");
103 fcntl_setfl(sigfd, O_NONBLOCK);
105 qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
112 static int qemu_signal_init(void)
118 static AioContext *qemu_aio_context;
119 static QEMUBH *qemu_notify_bh;
121 static void notify_event_cb(void *opaque)
123 /* No need to do anything; this bottom half is only used to
124 * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
128 AioContext *qemu_get_aio_context(void)
130 return qemu_aio_context;
133 void qemu_notify_event(void)
135 if (!qemu_aio_context) {
138 qemu_bh_schedule(qemu_notify_bh);
141 static GArray *gpollfds;
143 int qemu_init_main_loop(Error **errp)
147 Error *local_error = NULL;
151 ret = qemu_signal_init();
156 qemu_aio_context = aio_context_new(&local_error);
157 qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
158 if (!qemu_aio_context) {
159 error_propagate(errp, local_error);
162 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
163 src = aio_get_g_source(qemu_aio_context);
164 g_source_attach(src, NULL);
169 static int max_priority;
172 static int glib_pollfds_idx;
173 static int glib_n_poll_fds;
175 static void glib_pollfds_fill(int64_t *cur_timeout)
177 GMainContext *context = g_main_context_default();
182 g_main_context_prepare(context, &max_priority);
184 glib_pollfds_idx = gpollfds->len;
189 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
190 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
191 n = g_main_context_query(context, max_priority, &timeout, pfds,
193 } while (n != glib_n_poll_fds);
198 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
201 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
204 static void glib_pollfds_poll(void)
206 GMainContext *context = g_main_context_default();
207 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
209 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
210 g_main_context_dispatch(context);
214 #define MAX_MAIN_LOOP_SPIN (1000)
216 static int os_host_main_loop_wait(int64_t timeout)
219 static int spin_counter;
221 glib_pollfds_fill(&timeout);
223 /* If the I/O thread is very busy or we are incorrectly busy waiting in
224 * the I/O thread, this can lead to starvation of the BQL such that the
225 * VCPU threads never run. To make sure we can detect the later case,
226 * print a message to the screen. If we run into this condition, create
227 * a fake timeout in order to give the VCPU threads a chance to run.
229 if (!timeout && (spin_counter > MAX_MAIN_LOOP_SPIN)) {
230 static bool notified;
232 if (!notified && !qtest_enabled()) {
234 "main-loop: WARNING: I/O thread spun for %d iterations\n",
244 qemu_mutex_unlock_iothread();
249 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
252 qemu_mutex_lock_iothread();
259 /***********************************************************/
260 /* Polling handling */
262 typedef struct PollingEntry {
265 struct PollingEntry *next;
268 static PollingEntry *first_polling_entry;
270 int qemu_add_polling_cb(PollingFunc *func, void *opaque)
272 PollingEntry **ppe, *pe;
273 pe = g_malloc0(sizeof(PollingEntry));
276 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
281 void qemu_del_polling_cb(PollingFunc *func, void *opaque)
283 PollingEntry **ppe, *pe;
284 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
286 if (pe->func == func && pe->opaque == opaque) {
294 /***********************************************************/
295 /* Wait objects support */
296 typedef struct WaitObjects {
298 int revents[MAXIMUM_WAIT_OBJECTS + 1];
299 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
300 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
301 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
304 static WaitObjects wait_objects = {0};
306 int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
308 WaitObjects *w = &wait_objects;
309 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
312 w->events[w->num] = handle;
313 w->func[w->num] = func;
314 w->opaque[w->num] = opaque;
315 w->revents[w->num] = 0;
320 void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
323 WaitObjects *w = &wait_objects;
326 for (i = 0; i < w->num; i++) {
327 if (w->events[i] == handle) {
331 w->events[i] = w->events[i + 1];
332 w->func[i] = w->func[i + 1];
333 w->opaque[i] = w->opaque[i + 1];
334 w->revents[i] = w->revents[i + 1];
342 void qemu_fd_register(int fd)
344 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
345 FD_READ | FD_ACCEPT | FD_CLOSE |
346 FD_CONNECT | FD_WRITE | FD_OOB);
349 static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
355 for (i = 0; i < pollfds->len; i++) {
356 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
358 int events = pfd->events;
359 if (events & G_IO_IN) {
361 nfds = MAX(nfds, fd);
363 if (events & G_IO_OUT) {
365 nfds = MAX(nfds, fd);
367 if (events & G_IO_PRI) {
369 nfds = MAX(nfds, fd);
375 static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
376 fd_set *wfds, fd_set *xfds)
380 for (i = 0; i < pollfds->len; i++) {
381 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
385 if (FD_ISSET(fd, rfds)) {
388 if (FD_ISSET(fd, wfds)) {
391 if (FD_ISSET(fd, xfds)) {
394 pfd->revents = revents & pfd->events;
398 static int os_host_main_loop_wait(int64_t timeout)
400 GMainContext *context = g_main_context_default();
401 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
403 int g_poll_ret, ret, i, n_poll_fds;
405 WaitObjects *w = &wait_objects;
407 int64_t poll_timeout_ns;
408 static struct timeval tv0;
409 fd_set rfds, wfds, xfds;
412 /* XXX: need to suppress polling by better using win32 events */
414 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
415 ret |= pe->func(pe->opaque);
424 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
426 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
427 if (select_ret != 0) {
430 if (select_ret > 0) {
431 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
435 g_main_context_prepare(context, &max_priority);
436 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
437 poll_fds, ARRAY_SIZE(poll_fds));
438 g_assert(n_poll_fds <= ARRAY_SIZE(poll_fds));
440 for (i = 0; i < w->num; i++) {
441 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
442 poll_fds[n_poll_fds + i].events = G_IO_IN;
445 if (poll_timeout < 0) {
446 poll_timeout_ns = -1;
448 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
451 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
453 qemu_mutex_unlock_iothread();
454 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
456 qemu_mutex_lock_iothread();
457 if (g_poll_ret > 0) {
458 for (i = 0; i < w->num; i++) {
459 w->revents[i] = poll_fds[n_poll_fds + i].revents;
461 for (i = 0; i < w->num; i++) {
462 if (w->revents[i] && w->func[i]) {
463 w->func[i](w->opaque[i]);
468 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
469 g_main_context_dispatch(context);
472 return select_ret || g_poll_ret;
476 int main_loop_wait(int nonblocking)
479 uint32_t timeout = UINT32_MAX;
486 /* poll any events */
487 g_array_set_size(gpollfds, 0); /* reset for new iteration */
488 /* XXX: separate device handlers from system ones */
490 slirp_pollfds_fill(gpollfds, &timeout);
492 qemu_iohandler_fill(gpollfds);
494 if (timeout == UINT32_MAX) {
497 timeout_ns = (uint64_t)timeout * (int64_t)(SCALE_MS);
500 timeout_ns = qemu_soonest_timeout(timeout_ns,
501 timerlistgroup_deadline_ns(
504 ret = os_host_main_loop_wait(timeout_ns);
505 qemu_iohandler_poll(gpollfds, ret);
507 slirp_pollfds_poll(gpollfds, (ret < 0));
510 qemu_clock_run_all_timers();
515 /* Functions to operate on the main QEMU AioContext. */
517 QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
519 return aio_bh_new(qemu_aio_context, cb, opaque);